A family of visual generation models developed by Alibaba Cloud, along with a web service that lets anyone try them. Wan turns text prompts or still images into video with a cinematic look, handling complex motion, character animation, and even lip movement driven by audio. Since its first release in February 2025, its defining trait has been that the model weights are published as open source under the Apache 2.0 license — the Wan 2.1 and Wan 2.2 repositories on GitHub have each gathered more than ten thousand stars. In addition to the browser-based web app, iOS and Android apps are available, and commercial use through the cloud API is supported.
Key Features
- Video generation from text or images: Create video from a sentence such as “a person walking through a city at dusk,” or animate a single still image you already have. Wan also supports storyboard-driven workflows, keeping a consistent look across multiple cuts
- Expressive power from a Mixture-of-Experts architecture: The model splits processing across several internal “experts” with different roles (Mixture-of-Experts, or MoE), which helps it render complex motion — camera moves, subject action, character animation — more stably
- Audio-driven video generation: Supply audio and generate a person whose mouth movements and gestures follow it. This suits narrated explainer videos and avatar-style presentation
- Accurate text rendering in 12 languages: Wan is strong at drawing text inside video, rendering 12 languages without the usual distortion. That matters for footage containing logos, subtitles, or signage
- Color grading control: The color of the output can be adjusted precisely, so results can be matched to a brand’s visual identity or a project’s intended tone
- Open-source model weights: Models including Wan 2.1 and Wan 2.2 are published on GitHub under Apache 2.0, so you can run them on your own GPUs or fine-tune them further. Commercial use is permitted
- Multi-platform availability: Usable from a web browser as well as iOS and Android apps
Pricing
| Plan | Price | What it covers |
|---|---|---|
| Web app (wan.video) | Check the official site | Credit-based plans. Video and image generation from the browser |
| API (Alibaba Cloud Model Studio) | Pay-as-you-go (check the docs) | Billed by the duration and resolution of generated video. Commercial use allowed |
| Self-hosted (open-source release) | Free (Apache 2.0) | Run the model weights on your own GPUs. Infrastructure costs are separate |
Pricing is current as of August 2026. The web app’s pricing page is rendered dynamically, so no figures are quoted here. Please check the official pricing page and the Alibaba Cloud Model Studio documentation for the latest rates. Note that some third-party API resellers list a per-second rate for generated video, but those prices differ from the official ones.
Pros & Cons
✅ Pros
- Model weights are published under Apache 2.0, allowing self-hosting, fine-tuning, and commercial use with a high degree of freedom
- Multiple entry points — web app, mobile apps, and API — mean no complicated setup is needed just to try it
- Covers areas other models often struggle with, such as audio-driven generation and text rendering in 12 languages
- Includes production-oriented features like storyboard-based consistency and color grading control
- An active open-source community makes tooling and example work easy to find
⚠️ Cons
- The web app’s pricing is displayed dynamically on the official site, which makes it hard to estimate costs in advance
- Self-hosting requires substantial GPU memory and setup knowledge; it is not something a beginner can pick up casually
- Each generated clip has a length limit, so longer pieces must be produced in segments and joined
- Because the provider is a Chinese cloud company, some organizations will need to review data-handling policies first
- Artifacts common to all AI video generation — fingers, fine lettering — can still appear
Comparison with Similar Services
| Criteria | Wan | Sora | Runway | Kling AI |
|---|---|---|---|---|
| Provider | Alibaba Cloud (China) | OpenAI (US) | Runway (US) | Kuaishou (China) |
| Model release | Open source (Apache 2.0) | Closed | Closed | Closed |
| Self-hosting | Possible | Not possible | Not possible | Not possible |
| Main entry points | Web, iOS, Android, API | Web, app | Web, API | Web, app, API |
| Strengths | Openness, audio-driven generation, multilingual text rendering | Ease of use for general users | Production workflow including editing | Natural human motion |
Who Is It For
- Developers and researchers who want to run generative AI video on their own servers or GPUs
- Teams that want to customize the model to match their own visual style
- Anyone producing talking-head video — avatars or explainers — synchronized to narration
- Creators who need multilingual text, including Japanese, rendered cleanly inside video
- People who simply want to try text-to-video from a browser or phone first
Summary
Wan is Alibaba Cloud’s family of models aimed at cinematic-quality video generation, and its most distinctive trait is that the weights are released under Apache 2.0. You can pick the entry point that fits your goal: the web and mobile apps for casual use, the cloud API for embedding it into a product, and the open-source release for full control. On the other hand, the web app’s pricing has to be checked on the official site each time, and self-hosting demands real GPU resources. A practical path is to confirm the output quality within what you can try for free, then expand to the API or self-hosting once your use case is clear.